Daily Stock Forecasts optimizes and ranks machine learning models to predict the intraday movement of the stock market for the top 10 US Equities by Market Cap and a number of popular indicies.
Every trading day, DSF builds a number of classification models using historical candle+volume data. Each model's hyperparameters are optimized as well as the length of the lookback period per sample. Classification reprots are generated using test data. The f1 score is used to rank models.
Key files in the application hierarchy.
The frontend runs on a Google App Engine instance. It utilizes python, WebApp2, Jinja2 templating, JQuery, Google Charts, and soon Polymer and web components.
The backend and analysis can run locally if the datastore writing is disabled, but the current datastore exchange expects that the forecast is performed "inside the project" on a Google Compute Engine instance with the ability to securely access the Datastore.
Daily Stock Forecast was developed by Derek M Tishler,